Deep Random Search for Efficient Model Checking of Timed Automata

Abstract

Abstract. We present DRS (Deep Random Search), a new Las Vegas algorithm for model checking safety properties of timed automata. DRS explores the state space of the simulation graph of a timed automaton by performing random walks up to a prescribed depth. Nodes along these walks are then used to construct a random fringe, which is the starting point of additional deep random walks. The DRS algorithm is complete, and optimal to within a specified depth increment. Experimental results show that it is able to find extremely deep counter-examples for a number of benchmarks, outperforming Open-Kronos and Uppaal in the process.

Modeling, Verification, and Testing Using Timed and Hybrid Automata

“…To alleviate this problem different variants of "pure" random walk have been proposed. One such variant is the deep random search (DRS) algorithm proposed in [46] and applied in the context of timed automata. DRS stores during the random walk a subset of the nodes it visits, called a fringe, and then randomly backtracks to a node in the fringe when a deadlock (a node with no successors) is reached.…”

Resource-Aware Verification Using Randomized Exploration of Large State Spaces

“…In this case our algorithm URS reinitialize itself each time the memory is full. Note that in [14], the reinitialization of the algorithm DRS is not considered and the case of memory shortage is not studied. Here we place the two algorithms in the same context where re-initialization is applied each time the number of covered nodes reaches a prefixed threshold, which is, in our case, the memory size.…”

“…Also, URS is not performing a typical random walk, in the sense that it may choose to "branch" from different nodes along a random walk path. We have proposed comparison criteria such as mean cover time and used these to compare URS with a simplified version of the DRS algorithm proposed in [14]. We have also shown via experiments, that these two algorithms, when repeated several times, can explore a state space of more than 40% in addition to that explored by an exhaustive exploration based on breadth-first search.…”

“…The URS and SDRS algorithms [14] presents an extended random-walk based algorithm called Deep Random Search (DRS). The stop condition of DRS does not consider the limited memory size and supposes that the set of "non-closed" nodes (i.e., that have at least one non-visited successor) fits in main memory.…”

“…The re-initialization can be made from a random state of the previous walk and not necessary from the initial state. This has the advantage to minimize redundancy and reach deep states [14]. The local exhaustive search combined to random walk [15] explores better some regions of interest (dense regions for example) which cannot be well explored with only simple random walk.…”